1 / 54

Model Based Evaluation of Interactive Systems

Model Based Evaluation of Interactive Systems. Samit Bhattacharya Dept. of Computer Science & Engineering IIT Kharagpur. A Survey on UE Methods (Ivory & Hearst, 2001). Total UE methods surveyed = 132 The systems studied are of two types WIMP Web Based UI

shaun
Download Presentation

Model Based Evaluation of Interactive Systems

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Model Based Evaluation of Interactive Systems Samit Bhattacharya Dept. of Computer Science & Engineering IIT Kharagpur

  2. A Survey on UE Methods(Ivory & Hearst, 2001) • Total UE methods surveyed = 132 • The systems studied are of two types • WIMP • Web Based UI • The surveyed UE methods are classified into five method classes each involving a number of methods

  3. Survey Contd… Testing (an evaluator observes users interacting with an interface to determine usability problems) • Thinking-Aloud Protocol • Question-Asking Protocol • Shadowing Method • Coaching Method • Teaching Method • Co-discovery Learning • Performance Measurement • Log File Analysis • Retrospective Testing • Remote Testing

  4. Survey Contd… Inspection (an evaluator uses a set of criteria or heuristics to identify potential usability problems in an interface) • Guideline Review • Cognitive Walkthrough • Pluralistic Walkthrough • Heuristic Evaluation • Perspective-Based Inspection • Feature Inspection • Formal Usability Inspection • Consistency Inspection • Standards Inspection

  5. Survey Contd… Inquiry (users provide feedback on an interface via interviews, surveys, and the like) • Contextual Inquiry • Field Observation • Focus Groups • Interviews • Surveys • Questionnaires • Self-Reporting Logs • Screen Snapshots • User Feedback

  6. Facts and Figures • Each of the methods consists of three stages • Capture: data from user trials • Analysis: analysis of captured data • Critique: suggestion of possible improvements based on the analysis • Some statistics from the survey • The usage data is taken from the real users • 67% of the methods were executed manually • Automatic data capture in 13% cases • Automatic analysis in 18% of the methods • In 2% cases, the critique aspect had been automated

  7. Problems • Involving user may be very costly (e.g. pilots for designing flight simulators) • May not be possible to involve user at all (e.g. pilots, physically disabled) • Expert evaluators (not the designers) are needed to identify problems from usage data- a costly affair

  8. Way Out: Model Based Evaluation The Models Can Help in • Reducing the role of the user in testing • Automating the evaluation • Repetitive user trails can be dispensed with • Role of the expert evaluators can be reduced • A combination of the two

  9. Survey Revisited • Analytical Modeling - an evaluator employs user and interface models to generate usability predictions • Simulation - an evaluator employs user and interface models to mimic a user interacting with an interface and report the results of this interaction (e.g., simulated activities, errors, and other quantitative measures

  10. Remarks • Analytical modeling and simulation are engineering approaches to UE • They enable evaluators to predict usability with user and interface models • In this presentation, we will focus on such models only

  11. Two Modeling Paradigms • Modeling User (Computational Cognitive Models) • Modeling System (Formal/Mathematical Models)

  12. Computational Cognitive Models • Generic models of the user • Developed using theories and results from cognitive psychology

  13. Cognitive Psychology Theories in HCI • Informative/Prescriptive: used to develop design principles and guidelines • Predictive: used to predict behavior of interactive system. • Qualitative • Quantitative (Predictive Engineering Models)

  14. Predictive Engineering Models • Used to predict cognitive phenomena of interaction (e.g. Task completion time, memory load) • KLM • GOMS and variants (NGOMSL, CPM-GOMS) • PUM

  15. Programmable User Model • Proposed by Yong, Green and Simon (1989) • Idea was to develop a user model with embodied problem solving skills and common sense knowledge • Could be “programmed” with task knowledge • Essentially looks at the role that user knowledge plays in interaction • Much in the spirit of cognitive architecture without specific architectural components • No implementation exists

  16. Salient Points • All these models were developed in the 1980s • They were operational models based on principles • e.g. GOMS was based on the rationality principle and the problem space principle • Current practice is to base cognitive models on cognitive architectures

  17. Cognitive Architecture • “embodiment of a scientific hypothesis about those aspects of human cognition that are relatively constant over time and relatively independent of task” • Essentially integrative (cognition, perception, motor) theory of human information processing Cognitive Model = Architecture + Knowledge

  18. Characteristics of Architectures • Memory Structure • Perception (Visual, Auditory) • Motor action (Occulo-motor, manual) • Production system • Interaction mechanism (among architectural components) • Parameter values • Standard - system parameters fixed across all tasks (e.g. production cycle = 50 ms) • Typical - free to vary across task situations, but have more or less conventional values. (e.g. the time to recognize that a particular shape = 250 ms)

  19. Some Prominent Architectures in HCI • MHP (Model Human Processor) • ACT-R/PM (Adaptive Control of Thought-Rational/PerceptionMotor) • EPIC (Executive Process - Interactive Control) • Soar (originally stood for State Operators and Result) • ICS (Interactive Cognitive Subsystem)

  20. MHP

  21. MHP: Salient Points • Three interacting subsystems • Cognitive • Perception • Motor • Each with processor and memory • described by parameters • e.g., capacity, cycle time • Allows both serial and parallel processing • No implementation exists-used mainly for pedagogical purposes

  22. Soar

  23. Soar: Salient Points • Soar assumes behavior is a movement through problem space • Works based on the following principles: • Problem Space Principle: The rational activity in which people engage to solve a problem can be described in terms of (1) a set of states of knowledge, (2) operators for changing one state into another, (3) constraints on applying operators, and (4) control knowledge for deciding which operator to apply next • Principle of Rationality: People use knowledge to achieve goals

  24. Soar: Salient Points • Three types of long term memory • Procedural: procedural knowledge in the form of production rules • Declarative: facts about the world • Episodic: specific situations experienced before • Pervasive learning mechanism • Memory elements stored and retrieved in the form of Chunks • Perception and Motor modules not well developed

  25. Soar: Salient Points • Production system (i.e. decision making performed with production rules) • Decision cycle has five stages • Input: Working memory elements are created that reflect changes in perception • Elaboration: The contents of WM are matched against the “if” parts of the rules in LTM. All rules that match, fire in parallel, resulting in changes to the features and values of the state in addition to suggestions, or preferences, for selecting the current operator. As a result of the working memory changes, more rules may fire. Elaboration continues in parallel waves of rule firings until no more rules fire. • Decision: Decide which operator to choose next using preference • Application: Apply the operator to produce state transition • Output: Send commands to Perception/Motor interfaces

  26. EPIC

  27. EPIC: Salient Points • First of its kind to provide a detailed perceptual and motor mechanism • Basic assumptions • Production-rule cognitive processor • Parallel perceptual and motor processors • Multiple production rules can fire in a production cycle • No learning mechanism

  28. EPIC: Salient Points • Fixed architectural properties • Components, pathways, and most time parameters • Task-dependent properties • Cognitive processor production rules • Perceptual recoding • Response requirements and styles • Useful to model Motor-Perception intensive HCI tasks

  29. ACT-R/PM

  30. ACT-R/PM: Salient Points • A fusion of ACT-R and EPIC • ACT-R is the cognitive layer, production based • Unlike Soar and EPIC, can fire only one production at a time • If multiple production rules matches, an arbitration procedure called conflict resolution is used to determine the rule to fire • LTM = procedural + declarative

  31. ACT-R/PM: Salient Points • Information stored to and retrieved from the memory in Chunks • Each declarative memory elements has an activation value • Each production rule in the procedural memory has an utility value • These values are used for conflict resolution • Learning mechanism present but not pervasive like Soar

  32. ICS

  33. ICS: Salient Points • Represents cognitive activity as a configuration, or flow of information through different mental representations • Basic operation- transformation of a mental representation from one form, in which it describes a particular class of information about the world, into another form, in which it describes a different class of information

  34. ICS: Stages of Basic Operation • Storage - Store the incoming representation in a local form of LTM called the image record • Transformation - Access proceduralized knowledge acquired from previous experiences with the incoming information to try to produce the appropriate output representation • Procedural knowledge is embedded within the transformation processes • This embedded knowledge allows to produce an output for a particular input without accessing the image record • Revival - If procedural knowledge is poorly developed, ICS may try to access previous experiences that are similar to the incoming representation from the image record, to elaborate and refine it

  35. ICS: Salient Points • Stored information is said to be “revived by memory” rather than “retrieved from memory” because memory is seen as an active entity rather than a passive store • if the output representation that is produced is used successfully in task performance, the procedural knowledge relating the incoming and output representations is updated or strengthened, as appropriate

  36. ICS: Salient Points • The three basic operations—storage, transformation, and revival of information by memory—are grouped together within a “subsystem” dedicated to the processing of one particular class of information • Nine different sub systems with same structure

  37. ICS Sub System

  38. ICS: Subsystems • Sensory subsystems • VIS visual: encodes dimensions of light such as wavelength, brightness, user visual space • AC acoustic: encodes dimensions of sound such as frequency, timbre, intensity • BS body-state: encodes dimensions such as skeletal muscle tension • Structural subsystems • OBJ object: abstract structural description of entities and relations in visual space • MPL morpholexical: abstract structural description of entities and relations in sound space

  39. ICS: Subsystems • Meaning subsystems • PROP propositional: abstract description of entities and relations in semantic space • IMPLIC implicational: abstract description of human existential space abstracted over sensory and propositional input • Effector subsystems • ART articulatory: encodes dimensions such as force, target and timing of articulatory musculatures • LIM limb: encodes dimensions such as force, target, position and timing of skeletal musculatures

  40. Use of Cognitive Models in HCI • Explanation of (observed) interactive behavior • Evaluation of interactive systems • Cognitive Task Analysis (CTA) • Estimate human performance metrics from a task description • Compare interfaces with estimated values • Simulated Human User (SHU)/Synthetic User in simulated interaction environment

  41. Simulated Human User: Benefits • Can replace actual user, thus reducing the cost of prototype development and user trials • Can test a number of tasks with the interface • Allows the designer to test design alternatives/changes under many different conditions with operators of varying skill levels and other cognitive differences

  42. Formal Methods in HCI • Grammar Based (Linguistic) Formalism in the ’80s • BNF • Task-Action Grammar (TAG) • Linguistic Formalisms were developed mainly for Command Line Interface • Advent of Direct Manipulation Interfaces Inspired New Approaches

  43. State Transition Networks • Used to describe dialog (Interaction) • A number of variants were used • Basic STN • Generalized STN (GTN) • Augmented STN (ATN) • Not good at describing concurrent interaction (common in direct manipulation) • Alternate formalisms were used • Petri Nets • State charts

  44. Other Methods • Process Algebra • Temporal Logic • Markov Models • Matrix Algebra

  45. Remarks • All the early works were aimed to formally express relevant concepts (e.g. devices, dialogs) • Used to verify properties of interactive systems manually • Subsequent efforts were aimed at automatic verification

  46. Automatic Verification Deals with automatic verification of properties of interactive systems • Completeness: All possible transitions between all the states of interactions are taken care of. • Determinism: Same action cuases same state change always • Consistency: Action performed in different contexts does the same thing (difficult to verify formally) • Reachability: A user can get to a useful state from the current state easily. • Reversibility (undo): A special case of reachability

  47. Automatic Verification Contd… • Requires formal specifications of system and interaction • The choices to the designers are • Design custom specification language and suitable tools for verification • Use existing specification language and use available verification tools • Design custom specification language and mapper to existing specification languages • Each of the above choices have been tried

  48. Automatic Verification Contd… • Example of custom made specification language and tools • STN notation and HyperDoc • Matrix Algebra and MAUI • The second and third choice led to the use of model checkers for automatic verification

  49. Use of Model Checkers • Example of existing specification language and model checker • LOTOS  CADP • Example of custom made specification language and mapper: VEGPromela. The model checker used in this case is SPIN

  50. Comparison • Cognitive Models • Advantage: good at evaluating interfaces w.r.t. human factors (task completion time, memory load) • Disadvantage: Ignore system side of interaction (completeness, determinism, reachability, reversibility) • Formal Models (just the reverse)

More Related